Conservation tillage and the use of energy and other inputs in US agriculture

Energy Economics 20 Ž1998. 389]410

Conservation tillage and the use of energy and other inputs in US agriculture Noel D. Uri 1 ERS r RED r PMTB (Room 4056), US Department of Agriculture, 1800 M Street, NW, Washington DC 20036, USA

Abstract The effectiveness of conservation tillage practices in reducing the impact of agricultural production on the environment is dependent on what happens to energy, pesticide and fertilizer use as these practices are more extensively adopted. To gain some insight into this, the conservation tillage adoption decision is modelled. Starting with the assumption that the conservation tillage adoption decision is a two-step procedure } the first is the decision whether or not to adopt a conservation tillage production system and the second is the decision on the extent to which conservation tillage should be used } appropriate models of the Cragg and Heckman Ždominance. type are estimated. Based on farm-level data on corn production in the United States for 1987, the profile of a farm on which conservation tillage was adopted is that the cropland had above average slope and experienced above average rainfall, the farm was a cash grain enterprise, and it had an above average expenditure on pesticides and a below average expenditure on energy and a below average expenditure on custom pesticide applications. Additionally, for a farm adopting a no tillage production practice, an above average expenditure was made on fertilizer. Q 1998 Elsevier Science B.V. All rights reserved. JEL classification: R14 Keywords: Conservation tillage; Multiattribute utility; Agricultural production; Agricultural energy use

1 The views expressed are those of the author and do not necessarily represent the policies of the US Department of Agriculture or the views of other US Department of Agriculture staff members.

0140-9883r98r$19.00 Q 1998 Elsevier Science B.V. All rights reserved PII S0140-9883Ž97.00005-4

390

N.D. Uri r Energy Economics 20 (1998) 389]410

1. Introduction One approach to modelling the relationship between the adoption of conservation tillage practices 2 and the use of factor inputs including energy is to look at the farm level decision process as a multiattribute utility maximization analysis. Conventional utility theory of the Von Neumann and Morgenstern Ž1947. type assumes that an individual producer Žfarmer. makes decisions under certainty or under certainty equivalence ŽBaumol, 1972., with a single objective such as maximizing net farm income. Farmers now, however, are required to take into account aspects other than profit or net return above variable costs. For example, a farmer must also consider the externalities associated with his or her farming operation such as soil erosion and surface water and groundwater contamination. In this analysis it is assumed that a farmer is a maximizer of a multiattribute utility function. Multiattribute analysis offers a framework that permits a farmer to select among choices with different economic and environmental attributes. The complete analytics of this approach are not developed here. The interested reader is referred to Keeney and Raiffa Ž1976.. The multiattribute utility approach assumes that each farmer is a potential adopter of each different tillage practice. For conservation tillage practices, however, this may not be true. That is, irrespective of the potential economic and environmental impacts, conservation tillage might not be an appropriate cropping practice due to, for example, site-specific physical characteristics such as the slope of the cropland, the type of soil Žtexture and structure ., etc. ŽThorne and Thorne, 1979.. Pudney Ž1989. proposes modelling this sort of situation using discrete random preference regimes. This approach assumes that a farmer adopting conservation tillage has a different preference structure than a farmer not adopting such a practice. Thus, a zero observation reflects the decision not to use conservation tillage. Consequently, in the first stage of the model, a farmer decides whether or not to adopt conservation tillage. Non-adopters are then dropped from the sample. The second stage of the model focuses on conservation tillage adopters. Firstly, it is assumed that some farmers cannot be induced to adopt conservation tillage irrespective of its impact on net farm income or the environment. Again, Pudney’s discrete random preference regime is plausible, i.e. where conservation tillage adopters have a different preference structure than non-adopters. In this case, zero observations Žfarmer does not adopt conservation tillage. reflect the decision not to employ a specific conservation tillage practice and only adopters determine the underlying structure of the use of conservation tillage. A Heckman model, described subsequently, is an appropriate statistical model for implementing this theoretical approach ŽHeckman, 1979..

2 For the purpose of this study, conservation tillage is defined to be any tillage or planting system that maintains at least 30% of the soil covered by residue after planting to reduce soil erosion by water; or where soil erosion by wind is the primary concern, maintains at least 1000 pounds Žper acre. of flat, small grain residue equivalent on the surface during the critical wind erosion period ŽBull, 1993..

N.D. Uri r Energy Economics 20 (1998) 389]410

391

Alternatively, one can envision a farmer as evaluating his or her utility functions with and without the adoption of conservation tillage and then determining whether or not to adopt such a practice. This sequence is plausible if certain factors, such as environmental awareness andror concern among different farmers andror site-specific physical characteristics, relate directly to the quantitative distinction between adoption and non-adoption of conservation tillage and are independent of the extent to which conservation tillage is adopted by a farmer. One can model this situation by first assuming that an individual farmer’s utility function takes the form U s U Ž p Ž Py, x 1 , x 2 , . . . , x k < at . , Q Ž z1 , z 2 , . . . , z n ..

Ž1.

where p denotes the profit function, y denotes the production function, P denotes the price of output which is assumed to be given in the competitive market for the agricultural commodity 3 , x 1 , x 2 , . . . , x k denote the factors of production including energy and agricultural chemicals used in the production of the commodity, t denotes the adoption Žuse. of a conservation tillage practice, a is equal to one if a conservation tillage practice is adopted or has the potential to be adopted and zero otherwise, and QŽ z1 , z 2 , . . . , z n . represents the characteristics of the cropland and the farmer that serve to potentially influence the adoption of a conservation tillage practice. Note that UŽp , Q . is a monotonically increasing and concave Von Neumann-Morgenstern utility function. The profit function is given as

p s Py Ž x 1 , x 2 , . . . , x k < at . y c1 x 1 y c 2 x 2 y . . . yc k x k y V

Ž2.

where c1 , c2 , . . . , c k denote the stochastic imputed price of the factors of production, and V denotes the fixed cost of production. For the production function, ­ yr­ x k G 0 and ­ 2 yr­ 2 x k F 0 ŽAnderson et al., 1977.. Assuming the farmer has as his or her objective the maximization of expected utility which is a function of profits and conditional upon the adoption of conservation tillage and the characteristics of the cropland and the farmer, the problem becomes maximize E Ž U Ž p Ž Py, x 1 , x 2 , . . . , x k < at . , Q Ž z1 , z 2 , . . . , z n ..

Ž3.

where ­ Ur­ x i ) 0 and ­ 2 Ur­ 2 x i - 0.4 3

For ease of exposition, the production of just a single commodity is being considered. Extension of the analysis to the production of multiple commodities is straightforward ŽZilberman and Marra, 1993.. 4 These inequalities have economic interpretations. The first, ­ Ur­ x i ) 0, indicates that an increase in the use of factor of production x i will increase output y which in turn will increase profit and hence utility. The second inequality, ­ 2 Ur­ 2 x i - 0, used in conjunction with the first inequality indicates that while the increase in the use of a factor of production will increase output and hence profit and utility, the increase will be at a decreasing rate. That is, for a given factor of production, the production of the agricultural commodity and therefore profit and utility is characterized by diminishing marginal returns ŽStigler, 1966..

N.D. Uri r Energy Economics 20 (1998) 389]410

392

Substituting Ž2. into Ž3., the expected utility is E Ž U Ž Py Ž x 1 , x 2 , . . . , x k < at . y c1 x 1 y c 2 x 2 y . . . yc k x k y V . , Q Ž z1 , z 2 , . . . , z n .. . Ž4. The first order conditions for the maximization problem Ž3. with regard to a specific factor of production, x i , are E ŽŽ ­ U Ž p ,Q . r­ x i .Ž P Ž ­ y Ž at . r­ x i . y c i . s 0.

Ž5.

Since ­ UŽp , Q .r­ x i ) 0 by assumption, the optimal use of factor of production x i occurs when P Ž ­ y Ž at . r­ x i . s c i .

Ž6.

That is, the factor of production will be used up to the point where the value of the marginal product associated with that factor of production is just equal to its cost. This is a standard result from conventional neoclassical microeconomic theory ŽStigler, 1966.. The optimal use of the factor of production, however, will be conditioned by the tillage practice used because the production function is dependent on whether conventional tillage or conservation tillage is employed. Therefore, it is not possible to conclude a priori precisely what impact conservation tillage will have on input usage as the farmer endeavors to maximize profit-based expected utility that reflects the environmental externalities associated with agricultural production. In the context of an explicit adoption decision, a farmer will compare his or her utility at zero adoption with the utility at the level of conservation tillage adoption if it is decided to use one of these practices. The criterion for adoption is As

½

1 0

if Q - 0 otherwise

Ž7.

where UŽp U , Q . denotes the utility associated with the adoption of conservation tillage and UŽp , Q . denotes the utility associated with non-adoption.5 For the farmer who will not adopt conservation tillage under any circumstances, the indifference curves between conservation tillage adoption and non-adoption must be upward sloping and UŽp U , Q . y UŽp , Q . will be negative. Double hurdle Žor Cragg. models are statistical counterparts of this behavioral structure. 5

For ease of exposition just a single conservation tillage practice is being compared to conventional tillage. Obviously, the analysis can be expanded to include the comparison of the various conservation tillage practices and conventional tillage as well as a comparison among conservation tillage practices.

N.D. Uri r Energy Economics 20 (1998) 389]410

393

2. Statistical models The first model endeavors to associate the use of inputs, specific cropland characteristics, and a farmer’s characteristics with the decision to adopt conservation tillage. A probit model is used for this purpose. For all remaining models, the non-adopters are deleted from the sample. This implies that non-adopters are not at a standard corner solution. That is, changes in such things as the price of output and the prices of the factor inputs will not induce them to adopt conservation tillage. Attention is now focused on the econometrics of modelling the conservation tillage adoption decision and the extent of adoption decision. Firstly, for ease of exposition, it is assumed that both the adoption and extent of adoption equations are linear in parameters Ž b , g . with additive disturbance terms e and u, and the matrices V and W contain variables hypothesized to influence the adoption and extent of adoption decisions, respectively. Mathematically, the conservation tillage adoption specification is Q s b 9V q e for a s 1 if Q ) 0 and a s 0 otherwise.

Ž8.

Also, e ; nŽ0, 1. by assumption. The extent of conservation tillage adoption specification is

t s at UU

Ž 9a .

where

t UU s

0

½t

UU

if t U F 0 if t U ) 0

Ž 9b .

and

t U s g 9W q u. Note that it is assumed that u ; nŽ0, 1.. A positive adoption of conservation tillage t is observed only if a s 1 and t UU ) 0. Cragg or double-hurdle models postulate that to observe positive adoption, the farmer must pass two hurdles: Ž1. be a potential adopter of conservation tillage and Ž2. actually adopt conservation tillage ŽCragg, 1971; Lee and Maddala, 1985; Blundell and Meghir, 1987.. This allows for the possibility that zero conservation tillage adoption is a result of the extent of conservation tillage adoption decision. Hence, potential conservation tillage adopters may in fact not adopt any conservation tillage practice. Assuming correlated equation error terms allows for the possibility that the conservation tillage adoption and the extent of adoption decisions are made simultaneously: Ž e,u . ; n Ž 0,G .

Ž 10 .

N.D. Uri r Energy Economics 20 (1998) 389]410

394

where Gs

1

sr

sr

s2

where nŽ0, G . denotes a bivariate normal distribution with mean 0 and standard deviation G, r denotes the correlation coefficient, and s denotes the standard error. Using 0 to denote zero adoption of conservation tillage and q to denote positive adoption, the likelihood function for the dependent Cragg model is P w 1 y p Ž e ) yb 9V . p Ž u ) yg 9W < e ) yb 9V .x q

= P w p Ž e ) yb 9V . p Ž u ) yg 9W < y b 9V . g Ž t U < u ) yg 9W , e ) yb 9V .x q

Ž 11 . or P w 1 y F Ž b 9V ,g 9Wrs , r .x 0

= P F Ž b 9V . q srr Ž t U y g 9W . r 1 y r Ž 1rs .Ž f ŽŽ t U y g 9W . rs ..

'

q

Ž 12 . where p denotes the probability, F and f denote distribution and density functions respectively, and g Žv. s F Žv.rf Žv.. If the error terms e and u are independent Ži.e. r s 0., the independent Cragg model is obtained. This model assumes a feedback effect from the extent of conservation tillage adoption to the adoption decision ŽDeaton and Irish, 1986.. The Tobit model ŽMaddala, 1983. is a nested version of the independent Cragg model with F Ž b 9V . s 1. One advantage of the Cragg over the Tobit model is that the former allows variables to have differing effects on the adoption and extent of adoption decisions. A Heckman model assumes error terms of the adoption and extent of adoption equations are correlated and the adoption of conservation tillage decision dominates the extent of adoption decision. Domination implies that zero conservation tillage adoption is a result of the conservation tillage adoption decision and not the extent of conservation tillage adoption decision. Hence, only farmers with positive conservation tillage adoption levels are included in the extent of conservation tillage adoption decision. The model assumes the probability of a positive extent of conservation tillage adoption is equal to 1 given that a s 1 or pŽt U ) 0 < a s 1. and g Žt U
P wŽ p Ž e ) yb 9V .x Ž g Ž t U < u ) yg 9W .. q

Ž 13 .

N.D. Uri r Energy Economics 20 (1998) 389]410

395

with the log of the likelihood function written as S  y Ž 1r2. ln s 2 q ln Ž f ŽŽ t U y g 9W . rs ..

q

qr J ŽŽ t U y g 9W . rs . r Ž 1 y r . 4

'

q S ln Ž 1 y F Ž b 9V ..

Ž 14 .

0

where J s Fy1 Ž G . and G denotes the distribution function for u ŽLee and Maddala, 1985.. In contrast to the Cragg models, the Heckman model assumes that a farmer with no conservation tillage adoption provides no restrictions on the parameters of the extent of conservation tillage adoption equation. To see this, note that pŽ u ) yg 9W < e ) yb 9V . does not appear in the likelihood function for the Heckman model or is the expectation of t U , denoted by g, conditional upon e ) yb 9V. The Heckman model is simplified if the conservation tillage adoption and the extent of adoption equations are independent Ži.e. r s 0.. This model, termed the complete dominance model, separates into two independent components: Ž1. a probit for the adoption relationship and Ž2. ordinary least squares for the extent of conservation tillage adoption relationship using observations only on farmers who adopt conservation tillage ŽHeckman, 1979; Maddala, 1983..

3. Data and model specification The basic data used in estimating the conservation tillage adoption and extent of conservation tillage adoption relationships are for corn farms in the United States for 1987. The data come from the 1987 Farm Costs and Returns Survey ŽFCRS. conducted in February and March 1988 by the National Agricultural Statistics Service of the US Department of Agriculture. The FCRS is a stratified, multiframe survey consisting of a list frame and an area frame. The list frame farms were stratified by economic size, while area frame farms were stratified by use type. The survey is a full probability survey with all producers having a likelihood of being selected in the sample. Multiple versions of the FCRS are integrated into a single survey to simultaneously obtain data on farm organization, farm income, and expenses, assets and debt, and operator and household characteristics. Commodity-specific versions of the FCRS, which are conducted on a 4-year rotation, obtain data on enterprise production practices used in cost-of-production estimation. Data from all versions of the 1987 FCRS were used because corn was one of the commodities surveyed during that year. The sample of corn farms consisted of 1222 observations of which 825 were usable. Observations were deleted from the sample either because relevant information Žfor estimation purposes. was omitted or the data were seemingly incorrect.6 Note that the data on input use are in terms of expenditures and not physical quantities. It is not clear what magnitude of measurement error the use of this proxy introduces into the

396

N.D. Uri r Energy Economics 20 (1998) 389]410

analysis since the survey does not collect data on input prices.7 Moreover, the survey that does collect data on input prices is problematic ŽUri, 1994.. Thus, quantity data are simply not available. Merged with the FCRS data were specific topography, soil productivity, and soil texture data taken from the National Resource Inventory ŽSoil Conservation Service, 1982. which contains county-level data. For each county, the National Resource Inventory ŽNRI. sampled the physical characteristics of all non-federal rural land at several randomly selected points. Within county observations on soil texture, slope, and land capability class are quantified and averaged. The average includes only cropland observations. The NRI data are matched to a specific farm based on the county in which the farm is located. The topography variable, soil slope, measures the average cropland slope Ž%. for the county. Observations on soil texture are classified on a 5-point scale where 1 s sand, 2 s sandy loam, 3 s loam, 4 s clay loam, and 5 s clay. The numerical average for a county is then classified into one of three categories: sandy soil Žtexture F 2.3., loamy soil Ž2.3 - texture - 3.6., and clayey soil Žtexture G 3.6.. The sand and clay variables capture the soil texture effect relative to loam. The land capability classification system used in NRI classifies soils based on their ability to produce commonly cultivated crops. Land capability classes, identified 1 through 8, indicate progressively more limitations that restrict agricultural land use. For example, soils that are erosive, saline, shallow, stony, or wet limit land productivity. County observations on land capability are averaged and then classified such that land capability classifications less than 2.5 are defined as high productivity soils, while classifications greater than 3.5 are defined as low productivity soils. Finally, weather data including mean daily temperature averaged over June, July, and August and total monthly rainfall averaged over June, July, and August were collected from the National Oceanic and Atmospheric Administration ŽNOAA.. The data record for a specific farm was matched to the weather observation site nearest Žin a spatial sense. the farm. Note that 1987 was not an abnormal weather year. The mean average daily temperature and rainfall were not significantly different 8 than their historical averages. Two different types of conservation tillage 6

Incorrect data consisted of negative expenditures for factor inputs, acres treated with pesticides and on which fertilizer was applied being in excess of the number of acres planted, the number of irrigated acres being greater than the total number of acres planted, the seeding rate being unreasonably high Žin excess of 30 000 per acre., etc. 7 Combining expenditure data with price data would allow for the computation of quantity data. 8 That is, they did not depart by more than two standard deviations from their historical mean levels. 9 Under a no tillage cropping practice, the soil is left undisturbed before planting. Planting is completed in a narrow seedbed or slot created by a planter or drill. Weeds are controlled primarily with herbicides andror after planting cultivation ŽBull, 1993.. 10 With mulch tillage, the total surface is disturbed by tillage before planting. Tillage tools such as chisels or field cultivators Ždisks, sweeps, or blades. are used. Weeds are controlled with herbicides andror cultivation ŽBull, 1993..

N.D. Uri r Energy Economics 20 (1998) 389]410

397

are considered } no tillage 9 and mulch tillage.10,11 For the usable sample, 19% Ž159 of the 825 farms in the sample. of the farms used no tillage, 26% Ž218. used mulch tillage, while 4% Ž33. used both no tillage and mulch tillage on some portion of the cropland used for corn production. Presented in Table 1 are the variables and their definitions as used in the empirical models. Table 2 contains variable means and standard errors for the entire sample and for subsamples containing farmers who use only conventional tillage and farmers who have adopted conservation tillage Žno tillage and mulch tillage. on some portion of their cropland. Note that commonly used statistical methods for computing the mean and standard deviation are inappropriate in this study since the data were obtained from a complex survey ŽChamberlain, 1986.. Unlike simple random sampling, and as observed previously, the selection of an individual farm in the FCRS is not equally likely for all farms included in the list frame. Some farms have a higher probability of selection than others. Differences in the probability of selection introduces a bias to the conventional estimates of the mean and standard deviation ŽLee et al., 1989.. In order to overcome this problem, a weighted procedure in which the weights are equal to the inverse of the probability of selection must be used ŽFuller and Hidiroglou, 1978; Fuller, 1984.. Hypothesizing that a given variable is interrelated with the conservation tillage adoption decision and not the extent of adoption decision or vice versa is difficult. Consequently, both conservation tillage adoption and the extent of conservation tillage adoption are postulated to be functions of the various FCRS variables including expenditures on and use of factor inputs, farmerroperator characteristics, specific topography, soil productivity, and soil texture, and weather. Furthermore, because of the nature of the relationships hypothesized, it is important to realize that they are not meant to imply causality. Rather, they define identities that describe the relationships of conservation tillage adoption and the extent of conservation tillage adoption to the inputs used in corn production, farmerroperator characteristics, topography, soil productivity, etc. Finally, before presenting the estimation results, a few additional comments on the data and estimation are needed. Firstly, the estimation of the Cragg and Heckman models are carried out via maximum likelihood with the algorithm defined by Davidson, Fletcher, and Powell ŽDavidson and MacKinnon, 1993.. Next, the data that are farm-size-dependent Že.g. total fertilizer, pesticide, and energy expenditures, total yield, quantity of water applied, etc.. are divided by the number of acres planted to mitigate any potential effects of heteroscedasticity on the estimates due to farm size. Thirdly, since the temperature and rainfall for 1987 were not abnormal, use of the actual data as a proxy for expected temperature and precipitation12 is acceptable. Table 3a contains the coefficient estimates together with standard errors for the independent Cragg model and the complete dominance model for the no tillage 11

In 1994, 34.9% of the acreage planted in the United States employed some sort of conservation tillage practice. No-tillage was used on 12.5% and mulch tillage was used on 21.2% of planted acreage ŽEconomic Research Service, 1994..

N.D. Uri r Energy Economics 20 (1998) 389]410

398

production system and Table 3b contains coefficient estimates and standard errors of the estimates for mulch tillage.13 Also, as noted previously, because differences in the probability of selection introduces bias in the estimates of the coefficients

Table 1 Variable definitions Variable

Definition

FEXP PEXP

Per hectare a expenditures on fertilizer and soil conditioners Per hectare expenditures on insecticides, herbicides, fungicides, nematicides, and defoliants Per hectare expenditures on energy including diesel fuel, gasoline, liquified petroleum gas, and electricity Per hectare expenditures on hired labor Number of hours per hectare of ownerroperator labor devoted to corn production Per hectare fees paid for custom pesticide applications and services Proportion of planted hectares irrigated Per hectare quantity of water applied on irrigated hectareage Žin hectarerfeet . Per hectare seeding rate Proportion of hectares that received no pesticide treatment Per hectare corn yield Žkiloliter. The proportion of hectares idled under the acreage reduction programb The total number of hectares operated by the farmerroperator Defined to equal one if the farm is classified as a cash grain enterprise and zero otherwise Žoccurring, for example, if the farm is primarily operated to produce dairy products or beef or hogs. Defined to equal one if the farm is operated by a single owner and zero otherwise Defined to equal one if the farm is operated as a partnership and zero otherwise Defined to equal one if the farmerroperator graduated from college and zero otherwise Defined to equal one if the farmerroperator has some college but did not graduate and zero otherwise Defined to equal one if the farmerroperator has only a high school education and zero otherwise Defined to equal one if the farmerroperator is less than or equal to 30 years of age and zero otherwise Defined to equal one if the farmerroperator is more than 30 but less than or equal to 40 years of age and zero otherwise Defined to equal one if the farmerroperator is more than 40 but less than or equal to 50 years of age and zero otherwise Defined to equal one if the farmerroperator is more than 50 but less than or equal to 60 years of age and zero otherwise Average cropland slope Ž%. Defined to equal one if the texture is less than or equal to 2.3 and zero otherwise Žsee text for a discussion. Defined to equal one if the texture is greater than or equal to 3.6 and zero otherwise Žsee text for a discussion. Defined to equal one if the land capability classification is less than 2.5 and zero otherwise Žsee text for a discussion. Defined to equal one if the land capability classification is greater than 3.5 and zero otherwise Žsee text for a discussion.

FUEL LABOR OWNLABOR CUSTOM IRRIGATE WATER SEED NOPEST YIELD ARP TOTALHECT FARMTYPE

OWNTYPE1c OWNTYPE2 EDUCATION1 EDUCATION2 EDUCATION3 AGE1 AGE2 AGE3 AGE4d SLOPE TEXTURE1 TEXTURE2 SOILPROD1 SOILPROD2

N.D. Uri r Energy Economics 20 (1998) 389]410

399

Table 1 Ž Continued. Variable

Definition

AVGTEMP AVGRAIN PNT

Average monthly temperature for June, July, and August Ždegrees Fahrenheit . Average monthly rainfall for June, July, and August Žinches. Proportion of the total hectareage planted to corn on which no tillage has been adopted Proportion of the total hectareage planted to corn on which mulch tillage has been adopted

PMT a

This is per hectare of corn planted. See Lipton and Pollack Ž1989. for a technical definition of the hectareage reduction program. c Note that the variables associated with categorical data are not all inclusive. That is, a variable that corresponds to the observations not found in the enumerated variable categories is not defined. This is done to avoid the problem of singularity in the estimation. Thus, for example, no variable is defined for a farm ownership type Žsuch as a corporation. other than the one for a single owner and the one for a partnership. d The final category would be farmers who are more than 60 years of age. b

and their variances obtained via conventional estimation techniques, a weighted procedure in which the weights are equal to the inverse of the probability of selection must be used. Both of the models fit the data reasonably well for both no tillage and mulch tillage. The no tillage adoption equation correctly classifies Žas adopters or non-adopters. about 82% of the observations using the Ž0.5, 0.5. criterion while the mulch tillage correctly classifies about 86% of the observations. For this criterion, a correct classification means that the predicted probability of conservation tillage adoption is equal to or greater than 0.5 for an actual adopter and below 0.5 for a non-adopter ŽMaddala, 1983.. A maximum likelihood ratio test accepts the hypothesis that the independent Cragg model is an acceptable alternative to the Cragg model with dependence Žthe results are not presented but are available upon request. for both no tillage Ž x 2 s 1.01. and mulch tillage Ž x 2 s 2.32..14 This indicates that the conservation tillage adoption decision and the extent of conservation tillage adoption decision are not made simultaneously. A likelihood ratio test also indicates that the 12

Since the farmer does not know before making a conservation tillage decision what the weather Žtemperature and rainfall . will be like, he or she must base the decision on expectations, to the extent weather is a factor in the decision. The actual approach to expectations is not central to the current analysis. To minimize obfuscation, a common, straightforward approach to expectations formation is employed ŽIntrilligator, 1978.. 13 Note that in order to avoid problems in interpreting the results, farms that adopted mulch tillage exclusively are omitted from the sample used in the estimation of the no tillage models and those that adopted no tillage exclusively are omitted from the sample used in the estimation of the mulch tillage models. The impact of the 33 farms that adopted both no tillage and mulch tillage on some portion of their cropland is captured by introducing a proportion of cropland devoted to a specific conservation tillage practice variable in the extent of conservation tillage adoption relationship. 14 Both of the x 2 tests are performed with one degree of freedom. The critical value at the 5% level is 3.84.

400

N.D. Uri r Energy Economics 20 (1998) 389]410

complete dominance model is an acceptable form of the Heckman model Žthe results are not presented but are available upon request. for both no tillage Ž x 2 s 1.71. and mulch tillage Ž x 2 s 0.82..15 Since this likelihood ratio test is also a test for sample selection bias, the latter is not a significant problem for either no tillage or mulch tillage production systems ŽDhrymes, 1986.. Based on the results of the likelihood ratio tests, it is concluded that the independent Cragg and the complete dominance models are acceptable formulations for modelling the factors that are interrelated with the conservation tillage adoption decisions in the sample. Both of these models imply the conservation tillage adoption and the extent of adoption decisions are not made simultaneously. The major difference lies in the treatment of zero observations. The Cragg model implies that conservation tillage should be estimated over the entire population including both conservation tillage adopters and non-adopters and the complete dominance model implies the relevant population consists only of conservation tillage adopters. There are a comparable number of statistically significant coefficients Žat the 5% level. for both the Cragg and complete dominance models for both the conservation tillage adoption for the no tillage and the mulch tillage systems and the extent of adoption equations. Moreover, coefficient estimates, especially for those that are statistically significant, have similar signs across the two models. Note, however, making a direct comparison concerning such things as the order of magnitude of the coefficient estimates is difficult because the complete dominance model only includes conservation tillage adopters and the Cragg model uses all observations in the sample. Because a clear-cut decision cannot be made as to which model is preferable, the subjective preference is to choose the most general of the models } the Cragg. The Cragg model may be preferable because, unlike the complete dominance model, it assumes that a current conservation tillage non-adopter could be induced to adopt one or more of the conservation tillage practices if something changed. What do the results say about the relationship between conservation tillage adopters and the various FCRS variables including expenditures on energy and other factor inputs, farmerroperator characteristics, specific topography, soil productivity, and soil texture, and weather? Consider the no tillage production system first. No tillage adopters spend less on energy but more on fertilizer and pesticides than non-adopters. Energy expenditures are less under no tillage than they are under conventional tillage because there is a significant reduction in diesel fuel and gasoline used for tillage even though this is partially offset by the increase in energy use for the additional fertilizer and pesticide applications Žthis is discussed below. ŽUri and Day, 1992.. In a relative sense, energy used for tillage accounts for ; 11% of total direct farm energy use while fertilizer application accounts for 0.7% and pesticide application accounts for 0.8% ŽEconomic Research Service, 1987.. Thus, a reduction in farm energy use due to the reduction in or elimination of tillage would be expected to be substantially in excess of the increase in energy 15

Again both of the x 2 tests are performed with one degree of freedom.

N.D. Uri r Energy Economics 20 (1998) 389]410

401

Table 2 Sample means and standard deviations Žin parentheses . Variable

Full sample

Conventional tillage

No tillage

Mulch tillage

FEXP

$31.36 Ž$19.33. $12.54 Ž$8.25. $24.29 Ž$17.70. $9.85 Ž$13.60. 5.28 Ž6.51. $3.33 Ž$7.97. 0.07 Ž0.23. 1.06 Ž10.91. 58 271 Ž8434. 0.30 Ž0.10. 9.70 Ž2.74. 0.27 Ž0.19. 304.42 Ž336.33. 0.92 Ž0.05. 0.77 Ž0.41. 0.16 Ž0.37. 0.12 Ž0.32. 0.21 Ž0.41. 0.47 Ž0.49. 0.07 Ž0.27. 0.74 Ž0.25. 0.07 Ž0.21. 0.09 Ž0.19. 3.85 Ž2.76. 0.11 Ž0.30.

$29.70 Ž$19.60. $11.68 Ž$7.90. $24.84 Ž$18.19. $9.51 Ž$12.42. 5.65 Ž7.31. $3.70 Ž$8.37. 0.07 Ž0.24. 1.01 Ž10.30. 57 873 Ž8869. 0.32 Ž0.10. 9.60 Ž2.81. 0.27 Ž0.22. 294.92 Ž374.67. 0.91 Ž0.06. 0.78 Ž0.41. 0.17 Ž0.37. 0.10 Ž0.30. 0.19 Ž0.39. 0.49 Ž0.50. 0.08 Ž0.27. 0.75 Ž0.27. 0.08 Ž0.24. 0.06 Ž0.20. 3.58 Ž2.67. 0.11 Ž0.31.

$34.39 Ž$19.38. $13.73 Ž$8.29. $23.36 Ž$17.51. $10.02 Ž$13.65. 3.85 Ž3.46. $2.94 Ž$8.39. 0.04 Ž0.20. 0.40 Ž2.10. 58 868 Ž7308. 0.27 Ž0.07. 9.61 Ž2.77. 0.29 Ž0.20. 343.68 Ž289.18. 0.94 Ž0.07. 0.73 Ž0.45. 0.21 Ž0.41. 0.18 Ž0.39. 0.24 Ž0.43. 0.44 Ž0.50. 0.10 Ž0.30. 0.79 Ž0.26. 0.05 Ž0.23. 0.04 Ž0.22. 4.39 Ž2.99. 0.14 Ž0.35.

$33.01 Ž$18.52. $14.12 Ž$8.93. $23.06 Ž$15.82. $10.07 Ž$13.55. 5.26 Ž5.85. $2.49 Ž$6.12. 0.07 Ž0.24. 1.70 Ž14.72. 58 629 Ž8079. 0.30 Ž0.10. 10.04 Ž2.51. 0.32 Ž0.20. 310.89 Ž278.78. 0.95 Ž0.08. 0.80 Ž0.40. 0.13 Ž0.34. 0.12 Ž0.32. 0.29 Ž0.46. 0.46 Ž0.50. 0.08 Ž0.28. 0.81 Ž0.25. 0.06 Ž0.25. 0.04 Ž0.22. 4.28 Ž2.76. 0.08 Ž0.28.

PEXP FUEL LABOUR OWNLABOUR CUSTOM IRRIGATE WATER SEED NOPEST YIELD ARP TOTALHECT FARMTYPE OWNTYPE1 OWNTYPE2 EDUCATION1 EDUCATION2 EDUCATION3 AGE1 AGE2 AGE3 AGE4 SLOPE TEXTURE1

402

N.D. Uri r Energy Economics 20 (1998) 389]410

Table 2 Ž Continued. Variable TEXTURE2 SOILPROD1 SOILPROD2 AVGTEMP AVGRAIN PNT PMT Number of observations

Full sample 0.13 Ž0.33. 0.45 Ž0.50. 0.06 Ž0.24. 72.19 Ž3.29. 4.04 Ž1.41. 0.04 Ž0.10. 0.11 Ž0.23. 825

Conventional tillage 0.13 Ž0.34. 0.46 Ž0.50. 0.06 Ž0.23. 72.10 Ž3.67. 4.01 Ž1.53. 0.00 0.00 481

No tillage

Mulch tillage

0.07 Ž0.26. 0.36 Ž0.48. 0.07 Ž0.26. 73.13 Ž2.95. 4.56 Ž1.34. 0.19 Ž0.17. 0.07 Ž0.17. 159

0.16 Ž0.36. 0.46 Ž0.50. 0.07 Ž0.25. 71.79 Ž2.23. 3.73 Ž1.04. 0.02 Ž0.07. 0.42 Ž0.26. 218

Source: data for variables FEXP through AGE4 and PNT and PMT were taken directly from or computed based on data taken from the 1987 Farm Costs and Returns Survey, data for variables SLOPE through SOILPROD2 were taken from the National Resource Inventory, and AVGTEMP and AVGRAIN data were obtained from the National Oceanic and Atmospheric Administration. The text has relevant source citations.

use for fertilizer and pesticide applications. What is observed in the data is that, consistent with this expectation, a farmer who adopts no tillage is 16 percentage points more likely to have lower energy expenditures than a non-adopter. Moreover, for a no tillage adopter who uses less energy, a 1% increase in acreage under conservation tillage is associated with a 2.8% lower expenditure on energy. A farm that is a cash grain enterprise is about 24 percentage points more likely to adopt no tillage than, say, a dairy farm while the type of ownership Že.g. single owner versus a partnership. of the farm has no effect on the conservation tillage adoption decision. A farm with high productivity soil is about 22 percentage points less likely to adopt no tillage relative to a farm with average productivity soil while a farm with low productivity is about 12 percentage points more likely to adopt no tillage. The slope of the cropland is an important Žstatistically significant . factor associated with the adoption of no tillage. As the slope of the cropland increases, the propensity for water run-off and soil erosion increases 16 although the precise functional relationship is subject to debate ŽUri and Hyberg, 1990.. As noted previously, the no tillage production practice can be significant in reducing water run-off and soil erosion from fields. Farmers in the sample, on average, are using 16

This is reflected by the general use of the Universal Soil Loss Equation ŽUSLE. which contains a rainfall erosion index. The USLE was designed to capture the relationship between stream sediment loading and storm intensity ŽWischmeier and Smith, 1978..

N.D. Uri r Energy Economics 20 (1998) 389]410

403

this information. The estimation results suggest that a 1% increase in the slope of the cropland is associated with about a 44 percentage point increase in the likelihood that no tillage will be adopted.17 Average rainfall but not average temperature is associated with a small, but statistically significant Žat the 5% level. greater likelihood that no tillage is adopted. Thus, a 10% greater level of rainfall is associated with a 4-percentage point greater likelihood that no tillage is adopted on some portion of the cropland. This is additional confirmation that a farmer adopting no tillage is trying to mitigate water run-off and soil erosion since rainfall and soil erosion and, hence, stream sediment loading are inexorably intertwined ŽFawcett et al., 1994.. A farmer who adopts one conservation tillage practice is less likely to adopt another conservation tillage practice. That is, the farmer who adopts no tillage is 15 percentage points less likely to adopt mulch tillage. This result seems an anomaly since adoption of no tillage on a portion of the cropland does not inherently preclude adoption of mulch tillage on another portion of the cropland. The result and its attendant implications is an issue that needs to be explored further. A number of factors including expenditures on some inputs and farm and farm ownerroperator characteristics are found not to be associated with the adoption or non-adoption of the no tillage production practice. For example, the age and education level of the farmerroperator is not statistically significantly associated with the adoption of no tillage. Thus, the suggestion that younger, better educated farmers are more cognizant of the off-farm effects of agricultural production and hence are more inclined to adopt a conservation tillage practice is not borne out by the empirical results ŽOgg, 1992; Ferguson and Yee, 1995.. The texture of the soil, the total acres planted, the number of acres in the acreage reduction program, the extent of irrigation, and the proportion of acres not receiving any pesticide treatment are not interrelated with the adoption of no tillage. That is, such things as the size of the farm, the extent of government farm program participation, and the type of soil are unimportant in the conservation tillage adoption decision. This is surprising since the adoption of alternative farming systems has been shown to be intertwined with government farm program participation and farm size ŽDobbs et al., 1988; Ogg, 1990.. This issue is clearly deserving of further study. Hired labor and ownerroperator labor are two additional factors that are not significantly correlated with the conservation tillage adoption decision. That is, the decision to adopt no tillage farming has no relationship to the use of hired labor nor to the amount of labor expended by the farmerroperator. Also, no tillage farming is neither more nor less labor intensive than farming relying on conventional tillage practices ŽOffice of Technology Assessment, 1986.. Thus, it is not possible to conclude that no tillage is a labor saving production practice.

17

Note this value was computed based on the mean of the slope variable over the entire sample. Other response values are similarly computed. Complete computational details are available upon request.

404

N.D. Uri r Energy Economics 20 (1998) 389]410

Table 3 Parameter estimates of the independent Cragg model and the complete dominance model for Ža. no tillage and Žb. mulch tillage. Standard errors of the estimates are in parentheses Variable

Cragg model

Complete dominance

y7.315U Ž2.023. 0.041U Ž0.017. 0.029U Ž0.007. y0.012U Ž0.004. 0.011 Ž0.015. y0.107 Ž0.301. y2.811U Ž1.312. y0.001 Ž0.207. 0.251 Ž0.319. 0.003 Ž0.061. 0.415 Ž0.312. 0.003 Ž0.021. 0.396 Ž0.760. 0.001 Ž0.007. 0.461U Ž0.126. 0.251 Ž0.727. 0.413 Ž0.460. y0.944 Ž1.238. y0.611 Ž0.653. y0.369 Ž0.453. 0.116 Ž0.760. 0.073 Ž0.321. 0.054 Ž0.195. y0.041 Ž0.272. 0.092U Ž0.026. y0.078 Ž0.298. y0.233 Ž0.275. y0.333U Ž0.117. 0.171U Ž0.053. 0.055 Ž0.057. 0.153U Ž0.057.

y7.854U Ž2.178. 0.047U Ž0.021. 0.022U Ž0.010. y0.014U Ž0.006. 0.004 Ž0.063. y0.085U Ž0.033. y2.887U Ž0.706. y0.016 Ž0.037. 0.195 Ž0.402. 0.001 Ž0.317. 0.333 Ž1.760. 0.001 Ž0.022. 0.489 Ž0.731. 0.001 Ž0.074. 0.378U Ž0.134. 0.171 Ž0.276. 0.311 Ž0.581. y0.726 Ž1.184. y0.474 Ž0.727. y0.279 Ž0.391. 0.002 Ž0.210. 0.063 Ž0.251. y0.044 Ž0.189. 0.057 Ž0.231. 0.079U Ž0.027. y0.044 Ž0.315. y0.189 Ž0.271. y0.234U Ž0.113. 0.132U Ž0.047. 0.044 Ž0.037. 0.127U Ž0.050.

ŽB. Extent of conservation tillage adoption CONSTANT y0.222 Ž0.137. FEXP 0.078U Ž0.027. PEXP 0.059U Ž0.012. FUEL y0.009U Ž0.003. LABOR y0.004 Ž0.004. OWNLABOR y0.002 Ž0.014. CUSTOM 0.143U Ž0.062. IRRIGATE y0.005 Ž0.008. WATER 0.099 Ž0.376. SEED y0.006 Ž0.072. NOPEST 0.112 Ž0.342. YIELD 0.001 Ž0.051. ARP 0.039 Ž0.046. TOTALHECT y0.001 Ž0.014.

y0.376 Ž0.599. 0.071U Ž0.022. 0.062U Ž0.025. y0.015U Ž0.006. y0.009 Ž0.014. 0.054 Ž0.107. 0.293U Ž0.053. y0.013 Ž0.046. 0.073 Ž0.176. 0.011 Ž0.271. 0.027 Ž0.078. 0.005 Ž0.017. 0.119 Ž0.118. 0.001 Ž0.052.

Ža. ŽA. Conservation tillage adoption CONSTANT FEXP PEXP FUEL LABOR OWNLABOR CUSTOM IRRIGATE WATER SEED NOPEST YIELD ARP TOTALHECT FARMTYPE OWNTYPE1 OWNTYPE2 EDUCATION1 EDUCATION2 EDUCATION3 AGE1 AGE2 AGE3 AGE4 SLOPE TEXTURE1 TEXTURE2 SOILPROD1 SOILPROD2 AVGTEMP AVGRAIN

N.D. Uri r Energy Economics 20 (1998) 389]410 Table 3 Ž Continued. Variable

Cragg model

Complete dominance

FARMTYPE OWNTYPE1 OWNTYPE2 EDUCATION1 EDUCATION2 EDUCATION3 AGE1 AGE2 AGE3 AGE4 SLOPE TEXTURE1 TEXTURE2 SOILPROD1 SOILPROD2 AVGTEMP AVGRAIN PMT Sigma Log likelihood

0.016U Ž0.007. y0.018 Ž0.017. y0.027 Ž0.029. 0.017 Ž0.089. 0.018 Ž0.051. 0.012 Ž0.028. y0.087 Ž0.173. y0.003 Ž0.014. 0.009 Ž0.116. 0.030 Ž0.177. 0.043U Ž0.020. y0.007 Ž0.029. y0.005 Ž0.017. 0.002 Ž0.090. y0.021 Ž0.039. 0.002 Ž0.023. 0.009U Ž0.003. y0.051U Ž0.017. 1.717 Ž0.314. 252.2

0.009U Ž0.002. y0.035 Ž0.059. y0.161 Ž0.310. 0.044 Ž0.233. 0.072 Ž0.153. 0.076 Ž0.088. y0.037 Ž0.085. y0.014 Ž0.044. y0.041 Ž0.075. y0.026 Ž0.058. 0.077U Ž0.025. 0.049 Ž0.065. y0.018 Ž0.069. y0.007 Ž0.015. y0.086 Ž0.093. 0.004U Ž0.002. 0.005U Ž0.002. y0.197U Ž0.081. ŽA. q ŽB. s 267.6

Žb. ŽA. Conservation tillage adoption CONSTANT FEXP PEXP FUEL LABOR OWNLABOR CUSTOM IRRIGATE WATER SEED NOPEST YIELD ARP TOTALHECT FARMTYPE OWNTYPE1 OWNTYPE2 EDUCATION1 EDUCATION2 EDUCATION3 AGE1 AGE2 AGE3 AGE4 SLOPE TEXTURE1 TEXTURE2 SOILPROD1 SOILPROD2

y5.307U Ž2.011. 0.018 Ž0.037. 0.022U Ž0.009. y0.018U Ž0.007. 0.008 Ž0.053. y0.082 Ž0.092. y2.744U Ž1.144. y0.014 Ž0.038. 0.316 Ž0.520. 0.002 Ž0.073. 0.319 Ž0.427. 0.002 Ž0.025. 0.399 Ž0.689. 0.003 Ž0.042. 0.357U Ž0.117. 0.195 Ž0.633. 0.326 Ž0.517. y0.732 Ž1.064. y0.476 Ž0.756. y0.287 Ž0.538. 0.097 Ž0.326. 0.065 Ž0.167. y0.039 Ž0.188. 0.023 Ž0.208. 0.076U Ž0.025. y0.060 Ž0.281. y0.189 Ž0.250. y0.255U Ž0.128. 0.134U Ž0.046.

y0.021 Ž2.064. 0.001 Ž0.008. 0.045U Ž0.015. y0.009U Ž0.002. 0.007 Ž0.057. y0.054U Ž0.020. y2.554U Ž0.870. 0.001 Ž0.001. 0.207 Ž0.496. 0.001 Ž0.019. y0.610 Ž1.341. 0.001 Ž0.009. 0.611 Ž0.691. 0.001 Ž0.743. 0.248U Ž0.101. y0.070 Ž0.236. y0.336 Ž0.205. 1.243 Ž0.980. 1.140 Ž0.769. 0.518 Ž0.407. 0.013 Ž0.279. 0.212 Ž0.297. y0.165 Ž0.180. 0.075 Ž0.199. 0.138U Ž0.029. 0.105 Ž0.286. 0.338 Ž0.200. y0.240U Ž0.119. 0.056U Ž0.024.

405

406

N.D. Uri r Energy Economics 20 (1998) 389]410

Table 3 Ž Continued. Variable

Cragg model

AVGTEMP AVGRAIN

0.041 Ž0.030. 0.115U Ž0.051.

ŽB. Extent of conservation tillage adoption CONSTANT 0.015 Ž1.895. FEXP 0.002 Ž0.072. PEXP 0.038U Ž0.016. FUEL y0.007U Ž0.002. LABOR y0.006 Ž0.055. OWNLABOR y0.006 Ž0.019. CUSTOM y6.255U Ž2.592. IRRIGATE y0.013 Ž0.027. WATER 0.165 Ž0.388. SEED 0.001 Ž0.106. NOPEST y0.594 Ž1.370. YIELD 0.002 Ž0.003. ARP 0.566 Ž0.648. TOTALHECT 0.001 Ž0.065. FARMTYPE 0.243U Ž0.103. OWNTYPE1 y0.072 Ž0.215. OWNTYPE2 y0.303 Ž0.263. EDUCATION1 1.177 Ž1.766. EDUCATION2 1.066 Ž0.756. EDUCATION3 0.495 Ž0.387. AGE1 0.078 Ž0.200. AGE2 0.016 Ž0.254. AGE3 y0.180 Ž0.293. AGE4 0.065 Ž0.192. SLOPE 0.136U Ž0.028. TEXTURE1 0.101 Ž0.257. TEXTURE2 0.316 Ž0.294. SOILPROD1 y0.228U Ž0.109. SOILPROD2 0.502U Ž0.213. AVGTEMP 0.018 Ž0.022. AVGRAIN 0.193U Ž0.054. PNT y0.083U Ž0.004. Sigma 1.440 Ž0.281. Log likelihood 117.3 U

Complete dominance 0.029 Ž0.021. 0.207U Ž0.055. y1.008 Ž0.798. 0.036 Ž0.027. 0.033U Ž0.012. y0.042U Ž0.015. 0.003 Ž0.017. 0.012 Ž0.007. y2.793U Ž0.873. y0.016 Ž0.115. 0.111 Ž0.195. y0.015 Ž0.340. 0.809 Ž0.502. 0.008 Ž0.006. y0.009 Ž0.233. 0.002 Ž0.003. 0.139U Ž0.039. 0.025 Ž0.063. y0.015 Ž0.078. 0.311 Ž0.409. 0.259 Ž0.263. 0.127 Ž0.138. 0.018 Ž0.094. 0.032 Ž0.051. 0.109 Ž0.187. y0.047 Ž0.061. 0.147U Ž0.060. y0.073 Ž0.086. 0.100 Ž0.063. y0.147U Ž0.051. 0.411U Ž0.120. 0.023 Ž0.019. 0.036U Ž0.009. y0.073U Ž0.015. ŽA. q ŽB. s 109.5

Significant at the 5% level or better.

Table 3b presents information about the relationship between mulch tillage adopters and the various FCRS variables including expenditures on and use of factor inputs, farmerroperator characteristics, specific topography, soil productivity, and soil texture, and weather. The pattern of statistical significance of coefficient estimates of the variables is very similar to that observed for the no tillage production practice and the reasons are analogous. Consequently, an extensive discussion of the results is not required. Just the highlights will be noted. Mulch

N.D. Uri r Energy Economics 20 (1998) 389]410

407

tillage adopters spend less on energy but more on pesticides than non-adopters. Consistent with expectations, a farmer who adopts mulch tillage is 12 percentage points more likely to have lower energy expenditures than a non-adopter. Moreover, for a mulch tillage adopter who uses less energy, a 1% increase in acreage under conservation tillage is associated with a 1.1% lower expenditure on energy. A farm that is a cash grain enterprise is ; 18 percentage points more likely to adopt mulch tillage than, say, a beef or hog farm while the type of ownership of the farm has no effect on the conservation tillage adoption decision. A farm with high productivity soil is about 8 percentage points less likely to adopt mulch tillage relative to a farm with average productivity soil while a farm with low productivity is about 8 percentage points more likely to adopt mulch tillage. The slope of the cropland is a statistically significant Žat the 5% level. factor associated with the adoption of mulch tillage. The results suggest that a 1% increase in the slope of the cropland is associated with about a 25-percentage point increase in the likelihood that mulch tillage will be adopted. Average rainfall but not average temperature is associated with a statistically significant Žat the 5% level. greater likelihood that mulch tillage is adopted. A 10% greater level of rainfall is associated with a 42-percentage point greater likelihood that mulch tillage is adopted on some portion of the cropland. This effect is substantially greater than it is for the no tillage production practice results reported previously. The precise reason for the order of magnitude difference is properly a subject for future study. A farmer who adopts mulch tillage is 12 percentage points less likely to adopt no tillage. This inverse relationship is the same as that observed for no tillage. Clearly, the adoption of no tillage on a portion of cropland means that less cropland is available for mulch tillage but one should not necessarily witness a fall in the probability of the adoption of mulch tillage because of this. Why the adoption of one conservation tillage practice impacts the likelihood that another one will be adopted is an anomaly that is in need of further exploration. Several factors including expenditures on some inputs and farm and farm ownerroperator characteristics are found not to be associated with the adoption or non-adoption of the mulch tillage production practice. Thus, the age and education level of the farmerroperator is not statistically significantly associated with the adoption of mulch tillage. Consequently, as was the case for the no tillage production system, the suggestion that younger, better educated farmers are more aware of the environmental impacts of agricultural production than are their older, less well-educated counterparts and hence are more likely to adopt a conservation tillage practice is not borne out by the empirical results. The texture of the soil, the total acres planted, the number of acres in the acreage reduction program, the extent of irrigation, and the proportion of acres not receiving any pesticide treatment are not interrelated with the adoption of mulch tillage. That is, such things as the size of the farm, the extent of government farm program participation, and the type of soil are unimportant in the conservation tillage adoption decision.

408

N.D. Uri r Energy Economics 20 (1998) 389]410

Expenditures on hired labor and the amount of ownerroperator labor expended are additional factors not significantly interrelated with the conservation tillage adoption decision. That is, the decision to adopt mulch tillage farming has no relationship to expenditures on hired labor or the quantity of labor expended by the farmerroperator. Mulch tillage farming is neither more nor less labor intensive than farming relying on conventional tillage practices.

4. Conclusion Conservation tillage practices do have beneficial environmental effects in terms of reduced water run-off and mitigated soil erosion. There continues to be a question, however, as to their overall effectiveness in reducing the impact of agricultural production on the environment especially with regard to energy, pesticides and fertilizer. While it is generally recognized that pesticide use should increase under conservation tillage, what has not previously been adequately studied is the extent to which it will increase. Additionally, the impact of conservation tillage on fertilizer use is uncertain. To investigate these issues, it is assumed that the conservation tillage adoption decision is a two-step procedure } the first is the decision whether or not to adopt a conservation tillage production system and the second is the decision on the extent to which conservation tillage should be used. With this formulation of the problem, a double hurdle modelling approach is desirable. It is appropriate to use this approach and test whether decisions about conservation tillage adoption and the extent of adoption are separate, endogenous choices. Also, if it is credible that the conservation tillage adoption and extent of adoption decisions are made simultaneously then the equation error terms are correlated. Cragg and Heckman Žor dominance. models were estimated with and without the assumption of dependent error terms. The Cragg model with an independent error was the preferred model. Unlike the dominance models, the Cragg assumes all zero observations represent standard corner solutions. Based on farm-level data on corn production in the United States for 1987, the profile of a farm on which conservation tillage was adopted is that the cropland had above average slope and experienced above average rainfall, the farm was a cash grain enterprise, and it had an above average expenditure on pesticides and a below average expenditure on energy and a below average expenditure on custom pesticide application. Additionally, for a farm adopting a no tillage production practice, an above average expenditure was made on fertilizer.

References Anderson, J., Dillon, J., Hardarker, B., 1977. Agricultural Decision Analysis. The Iowa State University Press, Ames, IA. Baumol, W.J., 1972. Economic Theory and Operations Analysis. Prentice-Hall, Englewood Cliffs, NJ.

N.D. Uri r Energy Economics 20 (1998) 389]410

409

Blundell, R., Meghir, C., 1987. Bivariate alternative to the univariate tobit model. Journal of Econometrics 33, 179]200. Bull, L., 1993. Residue and tillage systems for field crops, SR AEGS 9310. Resources and Technology Division, Economic Research Service, US Department of Agriculture, Washington. Chamberlain, G., 1986. Panel data. In: Griliches, Z., Intriligator, M. ŽEds.., Handbook of Econometrics. North-Holland, Amsterdam. Cragg, J., 1971. Some statistical models for limited dependent variables with applications to the demand for durable goods. Econometrica 39, 829]844. Davidson, R., MacKinnon, J., 1993. Estimation and Inference in Econometrics. Oxford University Press, Oxford. Deaton, A., Irish, M., 1986. Statistical models for zero expenditures in household budgets. Journal of Public Economics 23, 59]80. Dhrymes, P., 1986. Limited dependent variables. In: Griliches, Z., Intriligator, M. ŽEds.., Handbook of Econometrics. North-Holland, Amsterdam. Dobbs, T., Leddy, M., Smolik, J., 1988. Factors influencing the economic potential for alternative farming systems. American Journal of Alternative Agriculture 8, 26]34. Economic Research Service, 1987. Energy and US Agriculture: State and National Fuel Use Tables, AGES861121. US Department of Agriculture, Washington. Economic Research Service, 1994. Agricultural Resources and Environmental Indicators, AH 705. US Department of Agriculture, Washington. Fawcett, R., Tierney, D., Christensen, B., 1994. Impact of conservation tillage on reducing runoff of pesticides into surface water. Journal of Soil and Water Conservation 49, 49]56. Ferguson, W., Yee, J., 1995. A logit model of cotton producer participation in professional scout programs. Journal of Sustainable Agriculture 5, 87]96. Fuller, W., 1984. Least squares and related analyses for complex survey design. Survey Methodology 10, 97]118. Fuller, W., Hidiroglou, M.A., 1978. Regression estimation after correcting for attenuation. Journal of the American Statistical Association 73, 99]104. Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47, 153]161. Intrilligator, M., 1978. Econometric Models, Techniques, and Applications. Prentice-Hall, Englewood Cliffs, NJ. Keeney, R., Raiffa, H., 1976. Decisions with Multiple Objectives. John Wiley, New York. Lee, E.S., Forthofer, R.N., Lerimor, R.J., 1989. Analyzing Complex Survey Data. Sage University Press, Newbury Park, CA. Lee, L., Maddala, G.S., 1985. The common structure of tests for selectivity bias, serial correlation, heteroscedasticity, and nonnormality in the Tobit model. International Economic Review 32, 238]251. Lipton, K., Pollack, S., 1989. A Glossary of Food and Agricultural Policy Terms, AIB-573, Economic Research Service, U.S. Department of Agriculture, Washington, DC. Maddala, G.S., 1983. Limited Dependent and Qualitative Variables in Econometrics. Cambridge University Press, Cambridge. Von Neumann, J., Morgenstern, O., 1947. Theory of Games and Economic Behavior, 2nd edn. Princeton University Press, Princeton, NJ. Office of Technology Assessment, 1986. Technology, Public Policy, and the Changing Structure of American Agriculture. Congress of the United States, Washington, DC. Ogg, C., 1990. Farm price distortions, chemical use, and the environment. Journal of Soil and Water Conservation 45, 45]48. Ogg, C., 1992. Addressing information needs to support sustainable agriculture policies. Journal of Sustainable Agriculture 2, 113]121. Pudney, S., 1989. Modelling Individual Choice: The Econometrics of Corners, Kinks, and Holes. Blackwell, London. Soil Conservation Service, 1982. National Resource Inventory. US Department of Agriculture, Washington. Stigler, G., 1966. The Theory of Price. Macmillan Publishing Company, New York.

410

N.D. Uri r Energy Economics 20 (1998) 389]410

Thorne, D.W., Thorne, M.D., 1979. Soil, Water and Crop Production. AVI, Westport, CT. Uri, N.D., 1994. Estimating the agricultural demand for natural gas and liquefied petroleum gas in the presence of measurement error in the data. International Journal of Energy Research 18, 783]794. Uri, N.D., Day, K., 1992. Energy efficiency, technological change, and the dieselization of American agriculture in the United States. Transportation Planning and Technology 16, 221]231. Uri, N.D., Hyberg, B., 1990. Stream sediment loading and rainfall } a look at the issue. Water, Air, and Soil Pollution 51, 95]104. Wischmeier, W., Smith, D., 1978. Predicting Rainfall Erosion Losses } A Guide to Conservation Planning, Agriculture Handbook No. 537. US Department of Agriculture, Washington. Zilberman, D., Marra, M., 1993. Agricultural externalities. In: Carlson, G., Zilberman, D., Miranowski, J. ŽEds.., Agricultural and Environmental Resources Economics. Oxford University Press, Oxford.